Five specialized agents. One accountable system.

OrynAI isn't one generalist prompt trying to do everything. It's five task-specific agents — Diagnose, Advise, Remediate, Validate, and Sequence — each grounded in your live instance and ServiceNow's official documentation, working together the way a Certified Technical Architect would: find the problem, explain it, fix it, and confirm the fix didn't break anything else.

What is OrynAI?

OrynAI is the AI system behind OrynIQ's platform-health scans, chat, and remediation. Most AI tools bolted onto ServiceNow are a single prompt with a long list of tools attached — accurate at finding things, inconsistent at explaining why they matter, and silent on whether a fix actually worked. OrynAI is built as five separate agents instead, each with a narrow job and only the tools that job needs.

A finding you can't understand is hard to justify fixing. A fix you can't confirm is safe will eventually break something. OrynAI's five agents exist to close both gaps — not just surface a list of problems, but quantify them, explain them, fix them, and prove the fix held.

The five agents

From finding to confirmed fix

Every phase is grounded in your live instance, your scan data, and ServiceNow's official documentation — never a guess dressed up as an answer.

01
Diagnose
Investigates your instance and quantifies what's wrong — not just that something is, but how much it affects and why.
02
Advise
Answers your questions conversationally, grounded in your instance and ServiceNow's official documentation — never memory dressed up as fact.
03
Remediate
Proposes a specific fix with the magnitude and reasoning attached. Nothing writes to your instance until you approve it.
04
ValidateIn development
Checks a proposed or applied fix against what actually depends on it — before it's presented as done.
05
SequenceIn development
Turns a pile of findings into a real roadmap — what to fix first, and why, based on what actually depends on what.

Diagnose — find it, and understand what it costs

A finding without scale attached is easy to ignore. Diagnose doesn't just flag an issue — it quantifies how many records, users, or CIs it touches, and states plainly whether your platform's trajectory is improving, stable, or degrading.

  1. 1

    Investigates your live instance

    Runs the compound analyzers relevant to what it's looking at, cross-referencing scan findings against real-time instance state — not a cached snapshot.

  2. 2

    Quantifies magnitude

    Every finding it surfaces carries an affected count, and where relevant, what percentage of your instance that represents — not just a severity label.

  3. 3

    States a clear trajectory

    Improving, stable, or degrading — with the specific findings driving that call, not a vague summary.

Advise — the education layer, on demand

A finding that can't be explained is hard to justify spending time on. Advise is where you ask "why does this matter" and get an answer grounded in your instance and ServiceNow's own documentation — not a training-data guess.

  1. 1

    Grounds every factual claim before stating it

    Role names, plugin behavior, ACL configuration, and encoded-query syntax are retrieved from ServiceNow's official documentation or read live from your instance — never stated from memory, which is where most ServiceNow AI tools get plausible-sounding details wrong.

  2. 2

    Cites its source

    "Per the Xanadu documentation..." or "Your instance shows..." — you know what's verified versus what's a standard-convention inference, every time.

  3. 3

    Stays in scope

    Advise answers ServiceNow platform-health questions — yours, your industry's, your peers' — and declines what isn't, rather than improvising outside its domain to seem helpful.

Remediate — propose, approve, apply

Every remediation proposal states what will change, how much it affects, and why it's the right fix — before you ever see an approve button. Nothing writes to your instance without your explicit sign-off.

  1. 1

    Proposes against an allowlist you control

    Every proposal is checked against a table/field allowlist your admin manages — script bodies, ACLs, schema tables, and security-critical user fields are never writable, by design, not by convention.

  2. 2

    States magnitude and rationale together

    A proposal isn't just "change field X to Y" — it carries the same quantified magnitude Diagnose surfaces, plus the grounded reasoning for why the change is correct.

  3. 3

    Applies only on your approval

    You approve, deny, or roll back — every action logged, every write immediately re-verified against your live instance.

Validate In development

A fix that can't be confirmed safe will eventually fail. Validate is the agent we're building to close that gap: after a fix is proposed or applied, it checks what actually depends on the thing that changed — not just whether the value we wrote is still there.

  1. 1

    Confirms the write held

    Re-reads the field after a change and confirms it matches what was written — the baseline check every remediation already gets today.

  2. 2

    Checks what depends on it

    Looks up what else references the thing that changed — other scripts that call it, CIs related to it, roles that chain through it, ACLs and business rules that key off it — and re-evaluates whether those are still in a valid state.

  3. 3

    Runs a regression test where one applies

    For script and form-affecting changes, triggers an ATF regression run against a non-production instance and folds the result into its verdict.

Sequence In development

A list of findings sorted by severity isn't a roadmap. Sequence is the agent we're building to turn a pile of validated findings into a real, dependency-aware plan — the kind of prioritization work a platform architect does by hand today.

  1. 1

    Finds the real dependencies

    Root-cause findings get fixed before the symptoms they cause. Fixes that would conflict with each other don't get scheduled into the same phase.

  2. 2

    Schedules against real constraints

    Severity and effort still matter as a tie-break — but the primary ordering comes from what actually has to happen first, not a fixed 30/60/90-day template.

  3. 3

    Explains every placement

    Every item in the plan comes with a narrated reason for its position — never a bare ordinal you have to take on faith.

What each agent is grounded in

Named mechanisms, not marketing language. This is what each agent actually checks against.

Diagnose

Compound analyzers + your live instance + official docs

Dozens of purpose-built analyzers cross-referenced against real-time instance state and ServiceNow's documentation corpus.

Advise

The same grounding, on demand, mid-conversation

Official documentation, your instance, and — when neither has the answer — ServiceNow's public knowledge base, always cited.

Remediate

A table/field allowlist your admin controls

Nothing OrynAI decides unilaterally is writable — script bodies, ACLs, schema, and security-critical fields are excluded by design.

Validate

The actual dependency graph of what changed

Script call chains, CI relationships, role and group containment, field references — plus an ATF regression run where one applies. Not just "re-read the value we wrote."

Sequence

Root-cause chains and conflict data your own instance produced

Real dependency edges, not a generic severity sort — built from findings and relationships specific to your platform.

See what OrynAI finds on your instance

Start a 30-day free trial and see Diagnose, Advise, and Remediate at work on your own data. No credit card, no sales call.

600+
automated checks feeding every diagnosis
30 days
free trial — no credit card required
0
writes to your instance without your explicit sign-off